JP4284265B2 - Moving picture coding apparatus, moving picture coding method, moving picture decoding apparatus, and moving picture decoding method - Google Patents

Description

  The present invention relates to a moving image encoding device, a moving image encoding method, a moving image decoding device, and a moving image decoding method that encode and decode an encoding target image with high accuracy.

  Conventionally, a method using motion compensated prediction is widely known as one of moving image encoding techniques. In a video encoding apparatus using motion compensated prediction, first, between an input image to be encoded and an image (local decoded image) obtained by decoding an already encoded image in the video encoding apparatus. Motion vectors are obtained. Next, motion compensation is performed using the obtained motion vector and the locally decoded image, and a predicted image for the input image is generated. The prediction error between the prediction image generated in this way and the input image is orthogonally transformed, and the orthogonal transformation coefficient is quantized and sent to the decoding apparatus together with the motion vector used for motion compensation prediction. . The decoding device receives the prediction error and the motion vector encoded by the encoding device in this way, generates a new prediction image using the decoded image that has already been decoded by the decoding device, The original image is decoded using the predicted image and the prediction error.

  In this way, in the moving image coding method for generating a predicted image by performing motion compensation prediction, in order to prevent degradation of the image quality of the decoded image with respect to the input image, the prediction error between the input image and the predicted image is reduced. There is a need.

As a method of reducing the prediction error, for example, first, a virtual pixel called a decimal point pixel generated using an interpolation filter is interpolated between pixels (integer pixels) originally present in the locally decoded image, and then A moving image in which a motion vector can be obtained with a more detailed resolution by obtaining a motion vector between a local decoded image (hereinafter referred to as an interpolated image) in which the decimal point pixels are interpolated and an input image. There is an image encoding method (see, for example, Non-Patent Document 1). Further, at this time, the filter coefficient of the interpolation filter for generating the decimal point pixel is adaptively changed with respect to the input image, so that the prediction error between the input image and the predicted image becomes smaller. A method for generating an interpolation filter for generating the above has also been proposed (for example, see Non-Patent Document 1).
T.A. Wedi, “Adaptive Interpolation Filter for Motion Compensated Prediction,” Proc. IEEE International Conference on Image Processing, Rochester, New York USA, September 2002

  As described above, according to the moving image encoding method disclosed in the conventional non-patent document 1, when the motion vector between the input image and the interpolated image points to the decimal point pixel of the interpolated image, By adaptively changing the interpolation filter according to the input image, the prediction error between the input image and the predicted image can be reduced.

  However, even when the local decoded image is interpolated by the decimal point pixels, the motion vector between the interpolated image and the input image is an integer pixel of the interpolated image (that is, a pixel originally present in the local decoded image). In the case of pointing, even if the interpolation filter is changed, the integer pixels of the interpolated image are not changed by this, so that the effect of reducing the prediction error between the input image and the predicted image cannot be obtained.

  The present invention has been made to solve the above-described problems of the prior art, and is an image obtained by motion compensation of an image obtained by filtering a locally decoded image (hereinafter referred to as a reference image). A filter for a locally decoded image is generated so that an error between the input image and the input image is reduced, and a prediction image is generated from a reference image obtained using the filter, thereby generating a filter between the input image and the prediction image. It is an object of the present invention to provide a moving picture coding method, a moving picture coding apparatus, a moving picture decoding method, and a moving picture decoding apparatus that can reduce a prediction error.

  In order to achieve the above object, the moving image encoding method of the present invention generates a predicted image using motion compensated prediction from a reference image obtained by filtering a locally decoded image, and the predicted image, the input image, In the moving picture coding method for encoding the quantized prediction error by orthogonally transforming and quantizing the prediction error between the input image and the locally decoded image, a first motion vector between the input image and the locally decoded image is obtained. Obtaining a filter for the locally decoded image such that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized; Filtering the locally decoded image with the filter to generate the reference image; and determining a second motion vector between the input image and the reference image. A step that is characterized by having the steps of: generating the predicted image by motion compensation by the reference picture and the second motion vector.

  In addition, the moving image encoding apparatus of the present invention generates a prediction image using motion compensated prediction from a reference image obtained by filtering a locally decoded image, and calculates a prediction error between the prediction image and the input image. In a moving image encoding apparatus that performs orthogonal transform and quantization to encode the quantized prediction error, a first motion vector between the input image and the locally decoded image, the input image, and the reference A motion detecting means for obtaining a second motion vector between the image and the input image so that an error between the image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized; Filter generating means for generating a filter for the locally decoded image, and reference image generating means for generating the reference image by filtering the locally decoded image with the filter Characterized in that it comprises a prediction image generation unit that generates the prediction image by motion compensation by the reference picture and the second motion vector.

  According to the present invention, a filter for a locally decoded image signal is generated so that an error between an image obtained by motion compensation of a reference image and an input image is small, and the reference image obtained by using this filter Since the prediction image is generated from the image, the prediction error between the input image and the prediction image can be reduced.

  Hereinafter, embodiments of the present invention will be described.

  FIG. 1 is a block diagram showing a moving picture coding apparatus according to an embodiment of the present invention.

  A moving image encoding apparatus according to this embodiment includes a subtractor 101 that generates a prediction error signal 12 from an input image signal 11 and a prediction image signal 16, an orthogonal transformer 102 that orthogonally transforms the prediction error signal 12, and an orthogonal transformation. The quantizer 103 that quantizes the orthogonal transform coefficient obtained by the multiplier 102, the inverse quantizer 104 that inversely quantizes the orthogonal transform coefficient quantized by the quantizer 103, and the inverse quantizer 104 An inverse orthogonal transformer 105 that reproduces a prediction error signal by performing inverse orthogonal transformation on the normalized orthogonal transform coefficient, and an adder that generates a locally decoded image signal 14 by adding the reproduced prediction error signal and the prediction image signal 16 106, a frame memory 107 for storing the local decoded image signal 14, and a local decoded image signal 15 read from the frame memory 107 and the input image signal 11 for motion compensation prediction. Generating a predictive image signal 16 Te includes a motion compensation predictor 108, the.

  FIG. 2 is a block diagram showing the configuration of the motion compensated predictor 108 according to the embodiment of the present invention.

  The motion compensated predictor 108 operates in conjunction with the switch 201 that switches the input destination of the locally decoded image signal 15, the reference image generator 202 that generates a reference image signal from the locally decoded image signal 15, and the switch 201. A switch 203 that switches a signal input to the detector 204, a motion detector 204 that obtains a motion vector from the local decoded image signal 15 or the reference image signal 11 and the input image signal 11 selected by the switch 203, the switch 201, and the switch In combination with 203, a switch 205 that switches the output destination of the motion vector obtained by the motion detector 204, the motion vector obtained by the motion detector 204, the local decoded image signal 15 and the input image signal 11, A filter generator 206 for generating a filter for the decoded image signal 15; A filter memory 207 for storing the filter generated by the filter generator 206, a subtractor 208 for calculating a difference between the filter generated by the filter generator 206 and the filter stored in the filter memory 207; A predicted image generator 209 that generates the predicted image signal 16 from the motion vector obtained by the detector 204 and the reference image signal generated by the reference image generator 202.

  Next, the operation of the moving picture coding apparatus according to the embodiment of the present invention will be described with reference to FIG. 1, FIG. 2, and FIG. FIG. 3 is a flowchart showing the operation of the video encoding apparatus according to the embodiment of the present invention.

  First, a moving image signal to be encoded is input to the moving image encoding device (step S101). Here, the moving image signal is composed of time-series still image data, and still image data at each time is input as an input image signal 11 to the moving image encoding apparatus. Hereinafter, still image data at each time is referred to as a frame.

  Next, in the subtracter 101, the difference between the pixel values of the corresponding pixels is calculated between the input image signal 11 and the predicted image signal 16 already generated in the motion compensated predictor 108, and the prediction error signal 12 is calculated. It is generated (step S102).

  The prediction error signal 12 is orthogonally transformed by the orthogonal transformer 102 (step S103), and the orthogonal transformation coefficient is quantized by the quantizer 103 (step S104). The orthogonal transform coefficient of the prediction error signal 12 quantized in this way is then input to the entropy encoder 109 and subjected to encoding processing.

  Further, the orthogonal transform coefficient of the quantized prediction error signal 12 is also input to the inverse quantizer 104 and is inversely quantized by the inverse quantizer 104 (step S105). Then, the inverse orthogonal transformer 105 performs inverse orthogonal transform to reproduce a prediction error signal (step S106).

  Next, the reproduced prediction error signal and the predicted image signal 16 input to the subtractor 101 in step S102 are added by the adder 106 to generate a local decoded image signal 14 (step S107), and the frame It is stored in the memory 107 (step S108).

  Next, the locally decoded image signal 15 is read from the frame memory 107 and input to the motion compensation predictor 108. Here, the locally decoded image signal 15 read out from the frame memory 107 may be determined in advance to use a locally decoded image signal of a past frame by a certain number of frames with respect to the current frame. It may be possible to designate a locally decoded image signal to be read out from. Further, the order of the frames of the input image signal to be processed is changed, and a locally decoded image signal of a future frame is first generated with respect to the currently processed frame and stored in the frame memory 107, and this is read out. You may enable it to use for the motion compensation prediction of the flame | frame currently processed.

  The motion compensated predictor 108 generates a predicted image signal 16 from the locally decoded image signal 15 and the input image signal 11 read from the frame memory 107 by using motion compensated prediction (step S109).

  Here, the operation of the motion compensation predictor 108 will be described with reference to FIGS. 2 and 4. FIG. 4 is a flowchart showing an operation for generating the predicted image signal 16 in the motion compensated predictor 108.

  First, in the motion compensation predictor 108, the states of the switch 201, the switch 203, and the switch 205 are initialized (step S201). That is, in the switch 201, the switch 203, and the switch 205, the terminal 201a, the terminal 203a, and the terminal 205a are in a conductive state, respectively.

  When the switch setting is initialized, the input image signal 11 and the locally decoded image signal 15 are input to the motion detector 204, and the motion detector 204 is configured to input the input image signal 11 and the locally decoded image signal 15. Motion vectors (hereinafter referred to as initial motion vectors) are calculated (step S202). As a method of calculating a motion vector from two image signals, for example, each image signal is divided into a plurality of regions (blocks), and the most similar block is searched from the two image signals for each block. A block matching method can be used in which the difference in position on the image signal is the motion vector of the block. In the block matching method, since motion vectors are obtained in units of blocks, the number of initial motion vectors is the same as the number of blocks. The initial motion vector thus detected by the motion detector 204 is then sent to the filter generator 206.

  The filter generator 206 uses the initial motion vector, the input image signal 11 and the local decoded image signal 15 to generate a filter for generating a reference image signal from the local decoded image signal 15 (step S203).

ここで、ｈ（ｉ，ｊ）は、図５に示すように、座標（ｘ＋ｉ，ｙ＋ｊ）に対するフィルタの重み係数、Ｎは線形和をとる画素の範囲を表す定数である。また、［ａ］は、実数ａを四捨五入した値を表す。 Here, as the filter generated by the filter generator 206, for example, for each integer pixel of the locally decoded image signal 15, a filter that takes a linear sum of pixel values of pixels within a certain range including the integer pixel, do it. That is, assuming that the pixel value of the pixel at the coordinates (x, y) of the locally decoded image signal 15 is S _L (x, y), the pixel value S _R of the pixel at the coordinates (x, y) of the corresponding reference image signal. It is assumed that (x, y) is obtained by equation (1).

Here, as shown in FIG. 5, h (i, j) is a filter weight coefficient for coordinates (x + i, y + j), and N is a constant representing the range of pixels that take a linear sum. [A] represents a value obtained by rounding off the real number a.

ここで、＞＞は右ビットシフトを意味する演算子を表す。また、（２）式では、ｍビットシフトして得られる値について四捨五入演算を行なうため、局部復号化画像信号１５の整数画素の画素値の線形和に２^ｍ−１を加算した後に、ビットシフトを行なっている。また、ｍはあらかじめ定めた定数である。 Further, in order to control the calculation amount for encoding instead of equation (1), using equation (2), the linear sum of the pixel values of integer pixels of the locally decoded image signal 15 is shifted to the right by m bits. It is also possible to use the value obtained by bit-shifting to the pixel value of the reference image.

Here, >> represents an operator that means a right bit shift. Further, in equation (2), since a rounding operation is performed on a value obtained by shifting m bits, a bit shift is performed after adding 2 ^m−1 to the linear sum of the pixel values of integer pixels of the locally decoded image signal 15. Is doing. M is a predetermined constant.

このように、オフセットｈ_{ｏｆｆｓｅｔ}を加算したものを参照画像信号の画素値とすることにより、画像全体の画素の平均的な輝度変化をも考慮したフィルタとすることができる。 Further, instead of the equation (1), the offset h is changed to a value obtained by bit-shifting the linear sum of the integer pixel values of the locally decoded image signal 15 to the right by m bits as in the equation (3). It is also possible to use the pixel value of the reference image as a result of adding _offset .

As described above, by adding the offset h _offset to the pixel value of the reference image signal, it is possible to obtain a filter that also considers the average luminance change of the pixels of the entire image.

なお、以下では、上述した（３）式を用いて参照画像信号を算出する場合について説明する。 Also, instead of equation (3), a value obtained by bit-shifting a value obtained by adding an offset to the linear sum of integer pixel values of the locally decoded image signal 15 to the right by m bits as in equation (4) The pixel value of the reference image may be used.

Hereinafter, a case where the reference image signal is calculated using the above-described equation (3) will be described.

ここで、Ｓ_Ｐ（ｘ，ｙ）は予測画像信号の座標（ｘ，ｙ）における画素値、ｖ_Ｉｉｘおよびｖ_Ｉｉｙは、それぞれ座標（ｘ，ｙ）が属するブロックｉの初期動きベクトルＶ_Ｉｉのｘ成分およびｙ成分を表す。 The filter generator 206 obtains a filter coefficient h (i, j) and an offset value h _offset of the filter for the locally decoded image signal 15. These filter coefficients h (i, j) and offset value h _offset are set so that the error between the predicted image signal obtained by motion compensation of the reference image signal with the initial motion vector and the input image signal 11 is minimized. Generated. The motion compensation of the reference image signal by the initial vector may be performed according to, for example, the equation (5).

Here, S _P (x, y) is the pixel value at the coordinates (x, y) of the predicted image signal, and v _Iix and v _Iiy are the initial motion vectors V _Ii of the block i to which the coordinates (x, y) belong, respectively. x component and y component are represented.

ここで、Ｓ（ｘ，ｙ）は、入力画像信号１１の座標（ｘ，ｙ）における画素値を表し、Σ_ｘ，ｙは、画像信号内に含まれるすべての画素についての和を表す。 Further, as the error between the predicted image signal and the input image signal 11, for example, a square error expressed by the equation (6) or an absolute value error expressed by the equation (7) can be used.

Here, S (x, y) represents the pixel value at the coordinates (x, y) of the input image signal 11, and Σ _{x, y} represents the sum for all the pixels included in the image signal.

The filter coefficient h (i, j) and the offset value h _offset that minimize the error between the predicted image signal obtained by the equation (6) or the equation (7) and the input image signal 11 are, for example, normal values of the least square method It can be obtained by solving the chemical equation. Alternatively, an approximation minimization method such as the Downhill Simplex method (see, for example, J. R. Nelder and R. Mead, “A simple method for function minimization,” Computer Journal, vol. 7, pp. 308-313, 1965). May be obtained as a filter that approximately minimizes the error.

The filters (h (i, j) and h _offset ) generated by the filter generator 206 in this way are sent to the reference image generator 202 and also sent to the filter memory 207 for storage. Further, the subtracter 208 calculates a difference between the filter generated by the filter generator 206 and the filter already stored in the filter memory 207, and generates a difference signal 17. Here, as a filter for calculating the difference between the filters, for example, a filter of a past frame for the current frame may be used.

ここで、ｈ_Ｍ（ｉ，ｊ）およびｈ_{Mｏｆｆｓｅｔ}は、それぞれフィルタメモリ２０７に記憶されている１フレームだけ過去のフィルタのフィルタ係数およびオフセット値である。 The calculation of the difference between the filters is performed by, for example, equation (8).

Here, h _M (i, j) and h _Moffset are the filter coefficient and the offset value of the previous filter by one frame stored in the filter memory 207, respectively.

  The difference signal 17 between the filters obtained in this way is sent to the entropy encoder 109 and encoded together with the orthogonal transform coefficient of the quantized prediction error signal 12. In this way, the filter coefficient and the offset value are not encoded as they are, but the difference between the filter that has already been generated and stored is taken, and the difference is encoded, thereby reducing the amount of information to be encoded. It becomes possible to reduce.

  When the filter is sent from the filter generator 206 to the reference image generator 202, the setting of the switch is changed (step S204). That is, in the switch 201, the switch 203, and the switch 205, the terminal 201b, the terminal 203b, and the terminal 205b are in a conductive state, respectively.

  When the switch setting is changed, the locally decoded image signal 15 is input to the reference image generator 202, and a reference image signal is generated (step S205). The reference image signal is generated by filtering the locally decoded image signal 15 according to the equation (3) using the filter sent from the filter generator 206. The reference image signal generated by the reference image generator 202 is then sent to the motion detector 204 via the switch 203.

  The motion detector 204 calculates a motion vector between the reference image signal sent from the reference image generator 202 and the input image signal 11 (step S206). As a motion vector calculation method, for example, the above-described block matching method may be used. The calculated motion vector is sent to the predicted image generator 209 via the switch 205. The calculated motion vector is also sent to the entropy encoder 109, and is encoded together with the orthogonal transform coefficient of the quantized prediction error signal 12 and the difference signal 17 between the filters sent from the subtracter 208.

ここで、ｖ_ｉｘおよびｖ_ｉｙは、それぞれ動き検出器２０４から送られる、座標（ｘ，ｙ）の属するブロックｉの動きベクトルＶ_ｉのｘ成分およびｙ成分を表す。 The predicted image generator 209 generates the predicted image signal 16 from the reference image signal sent from the reference image generator 202 and the motion vector sent from the motion detector 204 (step S207). The predicted image signal 16 can be obtained according to equation (9).

Here, v _ix and v _iy represent the x component and y component of the motion vector V _i of the block i to which the coordinate (x, y) belongs, respectively, sent from the motion detector 204.

  The prediction image signal 16 thus generated by the prediction image generator 209 is then sent to the subtractor 101 to generate a prediction error signal 12 with the newly input image signal 11. Used for.

  The operation of the motion compensation predictor 108 has been described above. As described above, the motion compensation predictor 108 generates a filter for the locally decoded image signal 15 for each frame, and generates a predicted image signal 16 for the input image signal 11 using this filter.

  Next, an orthogonal transform coefficient of the filter difference signal 17 for generating the reference image signal obtained by the motion compensation predictor 108, the motion vector 18, and the quantized prediction error signal 12 obtained by the quantizer 103 is The data is sent to the entropy encoder 109 and encoded (step S110). As the entropy encoder 109, for example, an arithmetic encoder may be used.

  These data encoded by the entropy encoder 109 are further multiplexed by the multiplexer 110 and output as encoded data 19 of the bit stream. The encoded data 19 is sent to a storage system or a transmission path (not shown).

  As described above, according to the video encoding apparatus according to the embodiment of the present invention, the filter for the integer pixels of the locally decoded image signal 15 is performed so that the error between the predicted image signal and the input image signal is reduced. The prediction error between the prediction image signal 16 and the input image signal 11 is reduced by generating the prediction image signal 16 from the reference image signal generated using this filter and the input image signal 11. Therefore, it is possible to prevent deterioration of the image quality of the decoded image signal with respect to the input image signal.

そして、ブロックごとに、（５）式によって求まる予測画像信号と入力画像信号１１との間の二乗誤差（（６）式）または絶対値誤差（（７）式）が最小になるように、フィルタの重み係数ｈ_ｋ（ｉ，ｊ）およびオフセットｈ_{ｋｏｆｆｓｅｔ}を定めればよい。このように、ブロックごとにフィルタを生成することにより、予測画像信号と入力画像信号の間の予測誤差をより小さくすることが可能になる。 In the above-described embodiment, the filter generator 206 generates a filter common to all integer pixels of the locally decoded image signal 15, but a different filter is generated for each block obtained by the motion detector 204. It is also possible. For example, the coordinate (x, y) of the reference image signal with the filter coefficient h _k (i, j) and the offset h _koffset for the coordinate (x, y) of the locally decoded image signal 15 belonging to the kth block. Assume that the pixel value at is obtained by equation (10).

Then, the filtering is performed so that the square error (Equation (6)) or the absolute value error (Equation (7)) between the predicted image signal obtained by Equation (5) and the input image signal 11 is minimized for each block. The weight coefficient h _k (i, j) and the offset h _koffset may be determined. Thus, by generating a filter for each block, it is possible to further reduce the prediction error between the predicted image signal and the input image signal.

  It is also possible to create a group by combining a plurality of blocks and generate one filter for each block unit. In this way, it is possible to reduce the prediction error between the predicted image signal and the input image signal, compared to the case where a filter common to all integer pixels of the locally decoded image signal 15 is generated. In addition, it is possible to reduce the amount of calculation for generating a filter as compared with the case of generating a filter for each block.

  In the above-described embodiment, the bit shift amount used in the expression (3) or (4) is a predetermined constant, but the bit shift amount can be made variable according to the encoding efficiency. It is also possible to encode this bit shift amount and transmit it to the decoder. Thus, by making the bit shift amount variable, it is possible to control the amount of information to be encoded efficiently.

  In the above-described embodiment, the locally decoded image signal 15 read from the frame memory 107 is a locally decoded image signal of a past frame for a predetermined time. The prediction image signal is generated by the flowchart shown in FIG. 4 for all the past or future local decoded image signals, and among them, the local decoding in which the prediction error between the prediction image signal and the input image signal is minimized. An image may be selected.

ここで、ｖ_ｋｘおよびｖ_ｋｙは、それぞれ動き検出器２０４から送られる、座標（ｘ，ｙ）の属するブロックｋの動きベクトルＶ_ｋのｘ成分およびｙ成分を表す。 In the embodiment described above, in step S207, the predicted image generator 209 generates the predicted image signal 16 using the reference image signal sent from the reference image generator 202 and the motion vector sent from the motion detector 204. However, the configuration of the motion compensated predictor 108 is changed so that the local decoded image signal 15 and the filter generated by the filter generator 206 are directly sent to the predicted image generator 209, so that the local decoded image 15, The predicted image signal 16 may be generated according to the equation (11) using the filter generated by the filter generator 206 and the motion vector sent from the motion detector 204.

Here, v _kx and v _ky represent the x component and y component of the motion vector V _k of the block k to which the coordinates (x, y) belong, respectively, sent from the motion detector 204.

  Next, a video decoding apparatus according to the embodiment of the present invention will be described.

  FIG. 6 is a block diagram showing a video decoding apparatus according to the embodiment of the present invention.

  A moving picture decoding apparatus according to this embodiment includes a demultiplexer 301 that separates encoded data 31, and an orthogonal transform of a quantized prediction error signal from the encoded data separated by the demultiplexer 301. An entropy decoder 302 that decodes a coefficient 32, a motion vector 33, and a difference signal 34 of a filter for generating a reference image signal, and an inverse quantization that inversely quantizes the orthogonal transform coefficient 32 of the quantized prediction error signal 303, an inverse orthogonal transformer 304 that reproduces the prediction error signal 35 by inverse orthogonal transformation of the orthogonal transformation coefficient of the prediction error signal, a frame memory 305 that stores a decoded image signal that has already been decoded, A reference image generator 306 for filtering the decoded image signal stored in the frame memory 305 to generate a reference image signal 36; and a reference image A prediction image generator 307 that generates a prediction image signal 37 from the reference image signal 36 generated by the generator 306 and the motion vector 33 sent from the entropy decoder 302, and a prediction image generated by the prediction image generator 307 An adder 308 that adds the signal 37 and the prediction error signal 35 reproduced by the inverse orthogonal transformer 304 to generate a decoded image signal, a filter memory 309 that stores the reproduced filter, and a filter memory 309 that stores the signal And an adder 310 that reproduces the filter by adding the filter difference signal 34 sent from the entropy decoder 302 to the reference image generator 306.

  Next, the operation of the moving picture decoding apparatus according to the embodiment of the present invention will be described using FIG. 6 and FIG. FIG. 7 is a flowchart showing the operation of the video decoding apparatus according to the embodiment of the present invention.

  First, as encoded data 31 to be decoded, encoded data 19 output from the moving image encoding apparatus in FIG. 1 is input to the moving image decoding apparatus shown in FIG. 6 via a storage system or a transmission system. (Step S301).

  The input encoded data 31 is separated by the demultiplexer 301 into encoded data of a motion vector, a differential signal of a filter for generating a reference image signal, and an orthogonal transform coefficient of a quantized prediction error signal. (Step S302).

Each separated encoded data is then sent to the entropy decoder 302 for decoding (step S303). An orthogonal transform coefficient 32 of the quantized prediction error signal decoded by the entropy decoder 302, a motion vector 33, and a difference signal 34 (Δh (i, j) and Δh _offset ) of a filter for generating a reference image signal. Are sent to the inverse quantizer 303, the predicted image generator 307, and the adder 310, respectively.

  The orthogonal transform coefficient 32 of the quantized prediction error signal is first inversely quantized by the inverse quantizer 303 (step S304), and then inversely orthogonally transformed by the inverse orthogonal transformer 304 to obtain the prediction error signal 35. Is reproduced (step S305).

ここで、フィルタの再生に用いられるフィルタメモリ３０９に記憶されているフィルタとしては、例えば、動画像符号化器において、現在フレームに対して１フレームだけ過去のフィルタを用いてフィルタ間の差分信号を生成した場合には、これに対応して、１フレームだけ過去のフレームをフィルタメモリ３０９から読み出して用いればよい。 The difference signal 34 (Δh (i, j) and Δh _offset ) of the filter for generating the reference image signal sent to the adder 310 is the filter (h _M (i, j)) stored in the filter memory 309. h _Moffset ) is added to reproduce the filter (h (i, j) and h _offset ) in the current frame (step S306). The filter may be regenerated according to the equation (12).

Here, as a filter stored in the filter memory 309 used for filter regeneration, for example, in a moving image encoder, a difference signal between filters is obtained by using a previous filter for one frame with respect to the current frame. If generated, the past frame corresponding to this may be read from the filter memory 309 and used.

  The filter reproduced by the adder 310 is sent to the reference image generator 306 and also sent to the filter memory 309 for storage.

ここで、Ｓ_Ｄ（ｘ，ｙ）は、フレームメモリ３０５に記憶されている復号化画像信号の座標（ｘ，ｙ）における画素の画素値を表す。また、読み出される復号化画像信号としては、例えば、動画像符号化装置において、現在フレームに対して一定のフレームだけ過去の局部復号化画像信号を用いて参照画像信号を生成した場合には、これに対応して、現在のフレームに対して、一定のフレームだけ過去の復号化画像信号をフレームメモリ３０５から読み出して用いればよい。 Next, the reference image generator 306 reads a past or future decoded image signal stored in the frame memory 305 for a certain time, performs filtering using a filter sent from the adder 310, and performs a reference image signal. 36 is generated (step S307). The reference image signal 36 is generated according to equation (13).

Here, S _D (x, y) represents the pixel value of the pixel at the coordinates (x, y) of the decoded image signal stored in the frame memory 305. In addition, as a decoded image signal to be read out, for example, when a reference image signal is generated using a past locally decoded image signal for a certain frame with respect to a current frame in a moving image encoding device, Corresponding to the current frame, the past decoded image signal for the current frame may be read from the frame memory 305 and used.

  The reference image signal 36 generated by the reference image generator 306 is then sent to the predicted image generator 307.

ここで、ｖ_ｉｘおよびｖ_ｉｙは、それぞれエントロピー復号化器３０２から送られる、座標（ｘ，ｙ）の属するブロックの動きベクトルＶ_ｉのｘ成分およびｙ成分を表す。 The predicted image generator 307 generates a predicted image signal 37 using the reference image signal 36 and the motion vector 33 sent from the entropy decoder 302 (step S308). The predicted image signal 37 is generated according to equation (14).

Here, v _ix and v _iy represent the x component and y component of the motion vector V _i of the block to which the coordinates (x, y) belong, respectively, sent from the entropy decoder 302.

  The prediction image signal 37 generated by the prediction image generator 307 is added to the prediction error signal 35 sent from the inverse orthogonal transformer 304 in the adder 308 to generate a decoded image signal (step S309). The time-series data of the decoded image signal generated in this way becomes a decoded moving image signal.

  The decoded image signal output from the adder 308 is also sent to and stored in the frame memory 305 (step S310).

  As described above, according to the video decoding device according to the embodiment of the present invention, the filter generated so that the error between the predicted image signal and the input image signal input to the video encoding device is reduced. Is used to generate a reference image signal 36, and a predicted image signal 37 is generated from the reference image signal 36. Therefore, deterioration of the image quality of the decoded image signal with respect to the input image signal input to the moving image encoding device is prevented. It becomes possible.

The block diagram which shows the structure of the moving image encoder concerning embodiment of this invention. The block diagram which shows the structure of the motion compensation predictor of the moving image encoder concerning embodiment of this invention. The flowchart which shows operation | movement of the moving image encoder concerning embodiment of this invention. The flowchart which shows operation | movement of the motion compensation predictor of the moving image encoder concerning embodiment of this invention. The figure showing the filter with respect to the local decoding image of embodiment of this invention. The block diagram which shows the structure of the moving image decoding apparatus concerning embodiment of this invention. The flowchart which shows operation | movement of the moving image decoding apparatus concerning embodiment of this invention.

Explanation of symbols

11... Input image signal 12, 35... Prediction error signal 13, 32... Orthogonal transform coefficient 14 and 15... Locally decoded image signal 16, 37. Prediction image signals 17, 34... Filter difference signals 18, 33... Motion vector 19, 31... Encoded data 36 .. Reference image signals 101, 208. Unit 102 ... Orthogonal transformer 103 ... Quantizer 104, 303 ... Inverse quantizer 105, 304 ... Inverse orthogonal transformers 106, 308, 310 ... Adders 107, 305, ... Frame memory 108 ... motion compensation predictor 109 ... entropy encoder 110 ... multiplexers 201, 203, 205 ... switches 202, 306 ... reference image generator 204 ... motion detection 206 ... filter generator 207,309 ... filter memory 209,307 ... prediction image generator 301 ... demultiplexer 302 ... entropy decoder

Claims (10)

A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In the moving picture coding method for coding the quantized prediction error,
Determining a first motion vector between the input image and the locally decoded image;
Generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Filtering the locally decoded image with the filter to generate the reference image;
Obtaining a second motion vector between the input image and the reference image;
Generating the predicted image by motion-compensating the reference image with the second motion vector;
A moving picture encoding method comprising: A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In the moving picture coding method for coding the quantized prediction error,
Determining a first motion vector between the input image and the locally decoded image;
Generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Filtering the locally decoded image with the filter to generate the reference image;
Obtaining a second motion vector between the input image and the reference image;
Filtering the image obtained by motion compensation of the locally decoded image with the second motion vector by the filter to generate the predicted image;
A moving picture encoding method comprising: A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In the moving picture coding method for coding the quantized prediction error,
Dividing the input image and the locally decoded image into blocks of a certain size, and obtaining a first motion vector between the input image and the locally decoded image in units of blocks;
Generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Filtering the locally decoded image with the filter to generate the reference image;
Dividing the input image and the reference image into blocks of a certain size, and obtaining a second motion vector between the input image and the reference image in units of blocks;
Generating the predicted image by performing motion compensation on the reference image using a second motion vector obtained in units of blocks;
A moving picture encoding method comprising: A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In the moving picture coding method for coding the quantized prediction error,
Dividing the input image and the locally decoded image into blocks of a certain size, and obtaining a first motion vector between the input image and the locally decoded image in units of blocks;
Generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Filtering the locally decoded image with the filter to generate the reference image;
Dividing the input image and the reference image into blocks of a certain size, and obtaining a second motion vector between the input image and the reference image in units of blocks;
Filtering the image obtained by motion compensation of the locally decoded image with the second motion vector obtained in units of blocks by the filter to generate the predicted image;
A moving picture encoding method comprising: The step of generating the filter generates, for each integer pixel of the locally decoded image, a filter for obtaining a weighted sum of pixel values of integer pixels in a certain range including the integer pixel. The moving image encoding method according to any one of claims 1 to 4. The step of generating the filter generates, for each integer pixel of the locally decoded image, a filter that bit-shifts a weighted sum of pixel values of integer pixels in a certain range including the integer pixels by a certain shift amount. The moving picture encoding method according to any one of claims 1 to 4, wherein: The step of generating the filter adds an offset to a value obtained by bit-shifting a weighted sum of pixel values of integer pixels in a certain range including the integer pixels by a certain shift amount for each integer pixel of the locally decoded image. 5. The moving picture encoding method according to claim 1, wherein a filter is generated. The step of generating the filter bit-shifts a value obtained by adding an offset to a weighted sum of pixel values of integer pixels in a certain range including the integer pixels for each integer pixel of the locally decoded image. 5. The moving picture encoding method according to claim 1, wherein a filter is generated. A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In a video encoding device that encodes the quantized prediction error,
Motion detection means for obtaining a first motion vector between the input image and the locally decoded image and a second motion vector between the input image and the reference image;
Filter generating means for generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Reference image generation means for generating the reference image by filtering the locally decoded image by the filter;
Predicted image generation means for generating a predicted image by performing motion compensation on the reference image with the second motion vector;
A moving picture encoding apparatus comprising: A prediction image is generated from the reference image obtained by filtering the locally decoded image using motion compensated prediction, and a prediction error between the prediction image and the input image is orthogonally transformed and quantized,
In a video encoding device that encodes the quantized prediction error,
The input image and the locally decoded image are divided into blocks of a certain size, a first motion vector between the input image and the locally decoded image is obtained in block units, and the input image and the reference A motion detection unit that divides the image into blocks of a certain size and obtains a second motion vector between the input image and the reference image in units of blocks;
Filter generating means for generating a filter for the locally decoded image so that an error between an image obtained by motion compensation of the reference image with the first motion vector and the input image is minimized;
Reference image generation means for generating the reference image by filtering the locally decoded image by the filter;
Predicted image generation means for generating the predicted image by performing motion compensation on the reference image using a second motion vector obtained in units of blocks;
A moving picture encoding apparatus comprising:

Images